[Cluster statistical analysis in epidemiology]

Statistical analysis represents a critical point in cluster analysis, because a methodology able to take into consideration the complexity of this analysis has not yet been developed. However, a common approach in statistical analysis of a suspected cluster is a necessary tool for public health operators who have to face population worries and requests. We propose an approach for the analysis of clusters and discuss the main limitations and strengths of the used methods. To this aim, we present, as a case study, the spatial clustering analysis of acute lymphoblastic leukaemia (ALL) cases among children in Rome between 2000 and 2011. Cases were selected through a record linkage of three different health and administrative current databases. Cases were geocoded at 3 spatial resolutions: 20 districts (D), 155 neighbourhoods (NB), and 5,812 census areas (CA). Indirect standardized incidence ratios (SIR) were computed for the NBs with Rome average incidence rate (IR) of ALL as reference and then smoothed by Besag-York-Mollie (BYM) model. General clustering was tested by Tango statistics, whereas localized clustering was detected through two different statistics: Besag and Newell's, and Kulldorf and Nagarwalla's. Both general and local clustering were tested at city level, using NBs as area units, and at district level, using CAs as area units. We identified 194 ALL cases in the 0-14 age group (IR: 43.7x1,000,000). SIRs ranged between 0.00 and 18.1 among NBs. After smoothing, a significant excess of cases was identified only in 3 Ds. At city level, no general clustering was highlighted (Tango's test p-value: 0.08), while both tests for local clustering were significant in one of the 3 Ds with the highest SIRs. Finally, at district level, although no general cluster was founded, a total of 7 clusters were identified in the 3 Ds with the highest SIRs, each cluster being composed by a number of cases ranging between 2 and 6. Results indicate the presence of clusters in some areas of Rome, which are evident only when the finest spatial resolution is used. This standardised procedure is an important tool to properly analyse potential clusters.